Compendium / Platform surfaces

Bing AI Performance Report

TypePlatform surface
Term maturityestablished
Operator maturityplausible
Lifecycleemerging
Relevanceoperational
Verified2026-10-28
The AI Performance report in Bing Webmaster Tools is a first-party count of how often Microsoft's AI answers cite a site's pages, broken down by page and by grounding query. It counts citations, not clicks, and it is a sample in which sparse activity may not surface, so it supports trends and like-for-like comparisons, not a full account of AI demand.

Consensus definition

The AI Performance report is a section of Bing Webmaster Tools that Microsoft announced as a public preview on 10 February 20261. It counts how often a site's content is cited as a source in Microsoft Copilot, in AI-generated summaries in Bing and in selected partner integrations1. The report shows Total Citations, Average Cited Pages, Grounding Queries, page-level citation activity and a timeline12. Average Cited Pages is the daily mean of unique pages cited, a measure of breadth, not a count of citations3. Grounding queries are grouped representations of the key phrases for which the site's content was referenced, not the full questions or prompts users typed. One query can map to several pages and one page to several queries314. From 16 June 2026 Microsoft began rolling out, in preview, intent labels, topic clusters, a Compare overlay against the previous period and Citation Share: the percentage of all citations shown for one grounding query that are attributed to the site53. On 13 October 2026 it added a beta tab, Topics Portfolio, that shows per topic the approximate citations across all sites, the approximate number of sites cited and queries where other sites are cited and the site is not. Microsoft calls these figures directional6. Microsoft describes the data as a sample of overall citation activity, not a complete log. Very low or infrequent citation activity may not surface, results may be refined as more data is processed, and totals may differ between the page, query and timeline views, partly because pages and queries may be sampled over slightly different windows3. Microsoft states that the report shows no traffic, clicks or engagement and no ranking or placement within an answer, and its metric list contains no impressions31.

rhinegold operator refinement

Position: the report is a sample, so each view answers a different question. Microsoft states that sparse citation activity may not surface and that totals can differ between the page, query and timeline views3. In our measurement work this shows up in the grounding-query export: in the weekly exports we checked, it lists no query below a small minimum number of citations, while the page export runs down to one or two citations. That is one reason the summed query rows stay below the page total. Microsoft names a second one, differing sampling windows3. Three rules follow. Use page-level citation totals for trends, because the page export reaches down to one or two citations, compared period on period in the same view and through the same export path. Use Citation Share only to follow your own share of one grounding query over time. It is a percentage within one query and shows neither who holds the rest nor a share across a topic35. Read the query list as a sample of the strongest grounding queries, not as the full demand behind your citations.

Operational use

Export the page, grounding-query and time-series data on a fixed weekly cadence. Microsoft refreshes the data daily with a short processing delay and offers history only for the windows in the dashboard (7 days, 30 days, 3 months, or a custom range within the available history)3, so a longer series exists only if you keep the exports. The dashboard filters by one grounding query or one page at a time. Combining a query with a page, or a topic with its pages, needs the CSV or Excel export of the filtered data3. Track three things separately. Page-level citation totals, per page and per page group, are the trend line. Citation Share is followed for the grounding queries that matter to you, each against its own history. The intent and topic labels describe which kinds of query cite you5. Write down the dashboard headline next to the summed export rows each time, because the two need not match3. Treat a step change on the timeline as a prompt to check what changed on the cited pages and in the data, not as proof of cause: Microsoft calls timeline changes observational3.

Measurement boundary

The report does not measure clicks, traffic or engagement from AI answers. A citation is a reference, not a visit31. It does not show where in an answer a page appears, whether it was the main source, or how it was presented31. It names no competing domains: Citation Share gives your percentage for a grounding query, not who holds the rest35. Grounding queries are grouped phrases, not prompts, so the report cannot tell you what users actually asked34. Citations are not conversations either: one practitioner analysis estimates, from the query variants it observed, three to five citations per conversation7. None of the Microsoft documents we read gives such a figure. Coverage is limited to Microsoft surfaces and unnamed partner integrations1. The report does not include Google's AI features, and one practitioner assumes that ChatGPT, Perplexity and Claude are not among the partners4. It says nothing about divergence across platforms. Because the data is a sample that may be refined later and sparse activity may not surface3, and Topics Portfolio leaves out topics with too little activity6, absence from the report does not show that a query or page was never cited.

Distinct from

From citation rate: citation rate is the share of AI answers in a prompt set you define that cite a brand, while the Bing report counts citations Microsoft recorded for your site, with no denominator of answers. From share of voice: share of voice divides your mentions by those of a defined competitor set across a prompt set. Citation Share divides your citations by all citations on one grounding query, names no competitors and does not add up to a share of a topic or market. Topics Portfolio's topic size counts citations across all sites, not searches6. From grounding: grounding is the retrieval mechanism, and the report counts only the sources shown as citations after retrieval, not every page retrieved. From grounding queries: that entry covers the queries themselves and how other providers expose them, while this one covers Microsoft's report. From grounded response rate: that rate asks how many answers retrieve sources at all, while the Bing report only sees answers that retrieved and cited your site. From Microsoft Copilot: Copilot is one of the surfaces the report covers, not the report itself.

A public example

One public practitioner analysis of a single site's export (91 days, 19,717 citations across 86 pages) found 69% of citations on one page and 90% on the top four, and daily citations falling by 97% from an early-December peak week to February7. It is one site, not a benchmark. It illustrates why a site-wide total can be driven by a handful of pages, and why a comparison that starts at a peak overstates a decline.

Common mistakes

  • Reading the summed grounding-query rows as total AI demand: the query list leaves out queries below a minimum number of citations, so it falls short of the page total3.
  • Comparing Citation Share across different grounding queries, or averaging it into a topic-level market share: it is a percentage within one query only35.
  • Treating citations as visits: the report carries no click, impression or traffic data31.
  • Comparing numbers taken from different views or filter paths, such as the dashboard headline against summed page rows, and reading the gap as a change: Microsoft says totals can differ by view and by sampling window3.
  • Attributing a timeline jump to a single page edit: Microsoft calls timeline changes observational, and citation patterns shift with user behaviour, model changes, freshness and partner refresh cycles35.

Sources & deeper reading

Last verified 2026-10-28 · Next review 2026-12-14
Question this measurementA metric can be right and the conclusion still wrong.Ask Dietrich →
Related terms
Cite this entry
rhinegold. “Bing AI Performance Report.” The Rhinegold Compendium. https://insights.rhinegold.de/compendium/bing-ai-performance-report/. Updated 2026-10-28.